Add disk-load's cpu_top tool: CPU by systemd unit and container (issue 741)

No tool named which units or processes used CPU outside containers during
the novox overload (issue 739). cpu_top reads each cgroup's cpu.stat (no
root needed), ranks units or containers busiest first, and says the
machine's CPU in use and the busy time no unit accounts for.

modules/disk-load/cmd/disk-load/cgroups.go

@@ -68,9 +68,10 @@ func (m *Machine) bottomDevices() map[string]bool {
 	return out
 }
 
-// unitIO is every unit's storage I/O, by its cgroup's path: the innermost .service or .scope, because
-// io.stat counts a cgroup's descendants too and an outer unit would count them twice.
-func (m *Machine) unitIO(counted map[string]bool) map[string]cgroupIO {
+// unitPaths is every systemd unit's cgroup path under /sys/fs/cgroup: the innermost .service or
+// .scope, because a cgroup's own stat file counts its descendants too and an outer unit would count
+// them twice. Shared by every reader that walks the cgroup tree (unitIO here, unitCPU in cpu.go).
+func (m *Machine) unitPaths() []string {
 	base := m.path("/sys/fs/cgroup")
 	var units []string
 	filepath.WalkDir(base, func(p string, d fs.DirEntry, err error) error {
@@ -89,11 +90,19 @@ func (m *Machine) unitIO(counted map[string]bool) map[string]cgroupIO {
 			outer[dir] = true
 		}
 	}
-	out := map[string]cgroupIO{}
+	out := make([]string, 0, len(units))
 	for _, u := range units {
-		if outer[u] {
-			continue
+		if !outer[u] {
+			out = append(out, u)
 		}
+	}
+	return out
+}
+
+// unitIO is every unit's storage I/O, by its cgroup's path.
+func (m *Machine) unitIO(counted map[string]bool) map[string]cgroupIO {
+	out := map[string]cgroupIO{}
+	for _, u := range m.unitPaths() {
 		text := m.read("/sys/fs/cgroup/" + u + "/io.stat")
 		if text == "" {
 			continue